Lesson 12 of 12 · All lessons in this module
  1. Fintech and the four Vs of Big Data
  2. Sources of Big Data and its challenges
  3. Financial data science and what makes financial data hard
  4. AI and how machine learning learns
  5. Splitting data for machine learning: size, leakage and time order
  6. Overfitting, underfitting and model fit
  7. Supervised, unsupervised and deep learning
  8. Reinforcement learning and the four learning approaches compared
  9. Inside a neural network: layers, backpropagation, GANs and VAEs
  10. Data science: processing and visualisation
  11. Text analytics, NLP and data science tools
  12. Large language models and generative AI in finance

This lesson is part of the 2027 curriculum. You are following the 2026 curriculum, where it is not taught in this form. Switch if you are sitting the exam under the 2027 curriculum.

Lesson 12 of 12 · 12 min

Large language models and generative AI in finance

LLMs are neural-network language models trained by self-supervised learning on general and financial text and then fine-tuned, so they can write, summarise and score sentiment; they are weak at exact numbers and need human oversight; generative AI more broadly learns patterns to create new, realistic content.

In short

  • Large language models (LLMs) use neural networks with transformer structures that track long-range context, so they generate coherent passages, not just the next word.
  • Financial LLMs depend on specialised corpora: they learn by self-supervised learning, first on general language, then on financial texts (annual reports, disclosures, earnings-call transcripts, market commentary).
  • After that they are fine-tuned (transfer learning) for tasks: text generation, summarisation, sentiment analysis and insight generation.
  • Risk: LLMs are trained to predict words, not to get numbers right, so errors in financial data are to be expected. Use human review or combine with data-driven models (a hybrid approach).
  • Generative AI creates new content that resembles real data (e.g., synthetic trading scenarios). Unlike traditional AI, which classifies or predicts, it generates; it is not Monte Carlo simulation or bootstrapping.

Unlock this lesson free for 7 days

Create a free account to get 7 days free: every lesson, video and flashcard, and the question bank. No card needed.

Start your 7-day free trialLog in